Official PyTorch implementation for Graph Matching based GNN Pre-Training [paper].

Yupeng Hou, Binbin Hu, Wayne Xin Zhao, Zhiqiang Zhang, Jun Zhou, Ji-Rong Wen. Neural Graph Matching for Pre-training Graph Neural Networks. SDM 2022.

python              3.7.7
pytorch             1.7.1
torch-geometric     1.6.3
cudatoolkit         10.1
rdkit               2020.09.1.0

Fine-tune with pre-trained GMPT-CL model on Bio

bash scripts/bio.sh

Pre-train from scratch

bash scripts/bio.sh pretrain

Check the results

python result_analysis.py --mode bio

For more detailed and customized usage, e.g., change datasets, GNN types, pre-trained models et al., please kindly refer to bio.sh and chem.sh.

Please refer to dataset-download to download Bio and Chem datasets.

Then the downloaded datasets should be moved to dataset/.

You can download the pre-trained GNN models from Google Drive and move them to bio_pretrain_model and chem_pretrain_model.

The implementation is reference to pretrain-gnns (backbone) and GraphCL (augmentation).

If you use this code for your research, please cite the following paper.

@inproceedings{hou2022gmpt,
  author = {Yupeng Hou and Binbin Hu and Wayne Xin Zhao and Zhiqiang Zhang and Jun Zhou and Ji-Rong Wen},
  title = {Neural Graph Matching for Pre-training Graph Neural Networks},
  booktitle = {{SDM}},
  year = {2022}
}